1780190 - Parent process deadlock when hiding a window

Status: NEW

(Core :: Widget: Gtk, defect, P3)

Description

[Emilio Cobos Álvarez (:emilio)]

Crash report: bp-5f20c0bf-74dc-4035-9827-9199b0220719

It seems the parent process is sending a sync Pause() to the compositor, but the compositor is sending a sync Resume() to the parent process.

Comment 1

[Sotaro Ikeda [:sotaro]]

I wonder if WindowSurfaceWaylandMB::Commit() might cause a dead lock.

First lock was held in

• https://searchfox.org/mozilla-central/rev/108c7843696fdf951083889d6fb858953139be96/widget/gtk/WindowSurfaceWaylandMultiBuffer.cpp#246

And second lock was tried to be held in

• https://searchfox.org/mozilla-central/rev/108c7843696fdf951083889d6fb858953139be96/widget/gtk/WindowSurfaceWaylandMultiBuffer.cpp#279

moz_container_wayland_add_initial_draw_callback() could call a callback if it is ready.
https://searchfox.org/mozilla-central/rev/108c7843696fdf951083889d6fb858953139be96/widget/gtk/MozContainerWayland.cpp#216

Comment 2

[Sotaro Ikeda [:sotaro]]

:stransky, can you comment to comment 1?

Comment 3

[Martin Stránský [:stransky] (ni? me) (PTO, back on Feb 17)]

(In reply to Sotaro Ikeda [:sotaro] from comment #1)

I wonder if WindowSurfaceWaylandMB::Commit() might cause a dead lock.

First lock was held in

• https://searchfox.org/mozilla-central/rev/108c7843696fdf951083889d6fb858953139be96/widget/gtk/WindowSurfaceWaylandMultiBuffer.cpp#246

And second lock was tried to be held in

• https://searchfox.org/mozilla-central/rev/108c7843696fdf951083889d6fb858953139be96/widget/gtk/WindowSurfaceWaylandMultiBuffer.cpp#279

moz_container_wayland_add_initial_draw_callback() could call a callback if it is ready.
https://searchfox.org/mozilla-central/rev/108c7843696fdf951083889d6fb858953139be96/widget/gtk/MozContainerWayland.cpp#216

Yes, that's correct - moz_container_wayland_add_initial_draw_callback() can fire the callback immediately so we have a deadlock here.

Comment 4

[Martin Stránský [:stransky] (ni? me) (PTO, back on Feb 17)]

Changed component according to comment 1.

Comment 5

[Martin Stránský [:stransky] (ni? me) (PTO, back on Feb 17)]

Bug 1780389 should handle the scenario from comment 1. I think the root cause here is the Pause()/Resume() call so let's fix that deadlock as extra bug.